NEA's Tiffany Luck: Enterprises Are Still Figuring Out Their AI ROI

Earlier this year, Silicon Valley was swept up in the trend of "tokenmaxxing," a practice where CEOs aggressively encouraged employees to push AI usage to its absolute limits. The goal was rapid integration and maximum productivity, but the strategy has quickly collided with financial reality. When the monthly cloud computing bills arrived, the enthusiasm for unrestricted AI experimentation cooled significantly.

The consequences of this unchecked adoption have already surfaced across major tech companies. Uber reportedly burned through its entire annual AI budget in just a few months, forcing a sudden and drastic reassessment of its deployment strategy. Elsewhere, some organizations have begun stripping Claude licenses from specific departments to curb spiraling costs, while Meta quietly dismantled its internal AI usage leaderboard, signaling a retreat from gamified, high-volume AI consumption.

This growing tension between aggressive AI integration and financial sustainability was the focal point of recent comments by Tiffany Luck, a partner at New Enterprise Associates (NEA). According to Luck, the current enterprise landscape is defined by a fundamental struggle: companies are still trying to figure out their AI return on investment (ROI).

The challenge is multifaceted. On one side, businesses face the undeniable pressure to adopt generative AI to remain competitive. On the other, they are grappling with the variable and often exorbitant costs associated with large language models, where expenses scale directly with token usage. When employees are incentivized to use AI for every task without understanding the underlying computational costs, the financial model breaks down.

Luck notes that the initial phase of AI deployment—characterized by blind enthusiasm and unlimited budgets—is officially over. Enterprises are now entering a more mature, pragmatic phase. This shift requires a rigorous evaluation of where AI actually drives measurable value versus where it merely acts as an expensive substitute for traditional computing or manual labor. Companies must transition from simply consuming tokens to strategically applying them.

Moving forward, Luck suggests that successful enterprise AI adoption will depend on developing tighter frameworks for usage, establishing clear cost controls, and identifying high-impact use cases that justify the expense. Until businesses can confidently draw a line from AI expenditure to tangible bottom-line improvements, the search for a sustainable AI ROI will remain the most critical challenge in the tech industry.